Human-mediated AI – the only model that actually works in classrooms

Ensuring learning is the starting point and keeping teachers as decision makers is essential for classroom AI use, explains Jérôme Nogues.

Teachers are not asking for more tools. They are asking much better questions.

• When should I use this?
• What does it replace?
• What does it risk?
• And quietly, underneath all of that, what happens to learning if I get this wrong?

That is the conversation we should be having.

Patterns of AI use

Over the past year, I keep seeing the same three patterns when it comes to how AI is adopted (or not adopted) in schools.

1. The first is a tool-first adoption approach. A school introduces a platform, runs some training and encourages staff to have a go. Some colleagues do interesting things, many just scratch the surface, others opt out entirely. Nothing really shifts and, if anything, the gap between classrooms widens.
2. The second is avoidance. AI is seen as a threat to thinking, academic integrity and the role of the teacher. So, access is restricted, policies are written and the message is clear: “Stay away from it”. Students, of course, do not. They just use it elsewhere, without guidance, structure or any real understanding of what they are doing. (FYI, neither of these first two approaches works).
3. The third one is quieter and less visible but far more powerful – ‘human-mediated AI’. The idea is simply that AI does not make decisions in the classroom, but teachers do. That sounds obvious but it isn’t, because we have to reframe our thinking. Instead of AI being the starting point, it’s the learning. The question becomes ‘What do I want my students to learn and where might AI support that, without doing the thinking for them?’

Where AI literacy makes the difference

AI is very good at producing answers but it’s not concerned with how those answers are reached. However, education is concerned about that and that’s where things get interesting. In practice, this plays out in two ways.

Firstly, teacher judgement. Yes, AI can save time, generate resources, draft questions and suggest ideas. But knowing what to keep, adapt or ignore is what matters. The teachers who use AI well are using it to help them spend more time where it matters – and that’s with their students.

Secondly, guided practice. Students can use AI to rehearse, try things out and build confidence. In languages, that might be speaking, experimenting with structures and getting immediate feedback.

Uncomfortable truths

This brings us to something slightly uncomfortable. Some of our tasks no longer work. Take the classic homework essay. For years, it has been a go-to way to develop ideas, structure thinking and practise writing. Now, it can be done in seconds. Fluent, coherent, impressive on the surface but empty in terms of learning. The response in many places has been to use AI detection tools. However, detection tools are unreliable, and, even when they work, they miss the point entirely.

If a task can be completed by AI without any real loss, then we must question the task itself. Human-mediated AI forces us to have that conversation. It pushes us towards tasks where students must explain, justify, adapt and reflect. Where, perhaps AI can support the process but not replace it. You could, for example, ask students to critique an AI response, improve it and explain why it falls short. We need to move towards making student practice more visible, understanding and appreciating how they got to their output.

Summary

AI  is not necessarily a bad thing because it gives us a chance to refocus on what matters. If AI is going to have a place in education, it will not be because it replaces teachers or magically reduces workload. It will be because it pushes us to design better learning, ask better questions and be more deliberate about what we value. And in that sense, there is nothing artificial about it at all.